Background Surgeon and patient treatment preferences are important threats to the internal and external validity of surgical trials such as PROFHER, which compared surgical versus non-surgical treatment for displaced fractures of the proximal humerus in adults. We explored the treatment preferences expressed by surgeons and patients in the trial and how these impacted on patient selection, trial conduct and patient outcome. Methods A series of exploratory secondary analyses of the PROFHER trial data were undertaken. We reviewed the extent of surgeon and patient treatment preferences (surgery or not surgery) at screening (n = 1250) as well as prior preference (including no preference) of randomised patients (n = 250), and assessed their impact on recruitment and adherence to follow-up and rehabilitation. Changes in treatment after 2 years’ follow-up were explored. Patient preference and characteristics associated with trial inclusion or treatment preference (t test, chi-squared test, Wilcoxon rank-sum test) were included as treatment interaction terms in the primary trial analysis of shoulder functioning (Oxford Shoulder Score, OSS). Results Surgeons excluded 17% of otherwise eligible patients based on lack of equipoise; these patients had less complex fractures (p < 0.001) and tended to be older (p = 0.062). Surgeons were more likely to recommend surgery for patients under 65 years of age (p = 0.059) and who had injured their right shoulder (p = 0.052). Over half of eligible patients (56%) did not consent to take part in the trial; these patients tended to be older (p = 0.022), with a preference for not surgery (74%; which was associated with older age, p = 0.039). There were no differential treatment effects (p value of interaction) for shoulder functioning (OSS) based on subgroups of patient preference (p = 0.751), age group (p = 0.264), fracture type (p = 0.954) and shoulder dominance (p = 0.850). Patients who were randomised to their preferred treatment had better follow-up rates (94 vs 84% at 2 years) and treatment adherence (90 vs 83% reported completing home exercises). Patients who were not randomised to their preferred treatment were more likely to change their treatment preference at 24 months (60 vs 26%). Conclusions The robustness of the PROFHER trial findings was confirmed against possible bias introduced by surgeon and patient preferences. The importance of collecting preference data is highlighted.